Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. An object detection method employing an Histograms of Oriented Gradients (HOG) feature (A) representing an intensity gradient of a target image for detecting existence of a target object in the target image based on an HOG feature (B) representing an intensity gradient computed in advance on a sample image capturing the target object, the object detection method comprising the steps of: computing a plurality of the HOG features (B) having different bin numbers for each of a plurality of local areas in the sample image to obtain a feature pattern representing the existence of the target object; constructing a classifier to judge the existence of the target object in the target image based on the feature pattern; and judging the existence of the target object in the target image by the classifier based on a plurality of the HOG features (A) having different bin numbers computed for each of a plurality of local areas in the target image.
This invention relates to object detection using Histograms of Oriented Gradients (HOG) features. The method addresses the challenge of accurately identifying target objects in images by leveraging multiple HOG features with varying bin numbers to improve detection robustness. The approach involves computing HOG features for both a sample image containing the target object and a target image where the object's presence is to be determined. For the sample image, multiple HOG features with different bin numbers are calculated for each of several local areas, forming a feature pattern that represents the target object's characteristics. A classifier is then trained using this feature pattern to distinguish the target object from other elements in the image. For object detection in the target image, HOG features with the same bin numbers as those used in the sample image are computed for corresponding local areas. The trained classifier evaluates these features to determine whether the target object exists in the target image. By using multiple HOG features with different bin numbers, the method enhances detection accuracy and reliability across varying image conditions.
2. The method according to claim 1 , wherein the target object is a human, and the classifier detects an entire body, an upper body, and a lower body of the target object in the target image based on the plurality of the HOG features (A) having the different bin numbers, detects each of directions of the entire body, the upper body, and the lower body of the detected target object, and judges a direction of the target object as a whole.
The object detection method described above, where Histograms of Oriented Gradients (HOG) are used to find objects, is adapted to detect humans. The classifier analyzes the HOG features, with different gradient orientations, to detect the entire body, upper body, and lower body of a person in the image. It also determines the direction each of these body parts is facing. Finally, it combines these directional estimations to determine the overall direction the person is facing.
3. The method according to claim 2 , wherein, from a plurality of bins of each of the HOG features (B), the bin effective to obtain the feature pattern is selected by a learning algorithm.
In the object detection method where Histograms of Oriented Gradients (HOG) are used, and a classifier is used to identify an object in an image, a learning algorithm selects the most important gradient orientations (bins) from the pre-computed HOG features of the sample image. This optimizes the feature pattern used to train the classifier, so it's more effective in finding the target object. From a plurality of bins of each of the pre-computed HOG features, the bin effective to obtain the feature pattern is selected by a learning algorithm.
4. The method according to claim 3 , wherein the learning algorithm is AdaBoost.
The object detection method previously described, where Histograms of Oriented Gradients (HOG) and a classifier are used with a learning algorithm selecting bins, uses AdaBoost as the learning algorithm to select the most relevant gradient orientations (bins) from the pre-computed HOG features. This optimized bin selection results in a more accurate classifier for object detection.
5. The method according to claim 1 , wherein, from a plurality of bins of each of the HOG features (B), the bin effective to obtain the feature pattern is selected by a learning algorithm.
In the object detection method where Histograms of Oriented Gradients (HOG) are used to find objects, a learning algorithm selects the most important gradient orientations (bins) from the pre-computed HOG features of the sample image. This optimizes the feature pattern used to train the classifier, so it's more effective in finding the target object. From a plurality of bins of each of the pre-computed HOG features, the bin effective to obtain the feature pattern is selected by a learning algorithm.
6. The method according to claim 5 , wherein the learning algorithm is AdaBoost.
The object detection method previously described, where Histograms of Oriented Gradients (HOG) and a classifier are used with a learning algorithm selecting bins, uses AdaBoost as the learning algorithm to select the most relevant gradient orientations (bins) from the pre-computed HOG features. This optimized bin selection results in a more accurate classifier for object detection.
7. An object detector employing an Histograms of Oriented Gradients (HOG) feature (A) representing an intensity gradient of a target image for detecting existence of a target object in the target image based on an HOG feature (B) representing an intensity gradient computed in advance on a sample image capturing the target object, the object detector comprising: a calculator computing a plurality of the HOG features (B) having different bin numbers for each of a plurality of local areas in the sample image to obtain a feature pattern representing the existence of the target object, and further computing a plurality of the HOG features (A) having different bin numbers for each of a plurality of local areas in the target image; and a classifier constructed by the calculator based on the feature pattern, and judging the existence of the target object in the target image based on the plurality of the HOG features (A) having the different bin numbers computed by the calculator.
An object detector uses Histograms of Oriented Gradients (HOG) to find objects in images. It includes a calculator that first computes HOG features for a sample image containing the target object. This pre-computed HOG data, with varying gradient orientations (bin numbers) for small local regions (cells) in the sample image, creates a feature pattern representing the object. Then, the calculator computes HOG features for a target image (again with different gradient orientations for each cell). A classifier is built by the calculator using the sample image feature pattern. Finally, the classifier uses the HOG features from the target image to determine if the object exists in that image.
8. The object detector according to claim 7 , wherein the target object is a human, and the classifier detects an entire body, an upper body, and a lower body of the target object in the target image based on the plurality of the HOG features (A) having the different bin numbers, detects each of directions of the entire body, the upper body, and the lower body of the detected target object, and judges a direction of the target object as a whole.
The object detector described above, where Histograms of Oriented Gradients (HOG) are used to find objects, is adapted to detect humans. The classifier analyzes the HOG features, with different gradient orientations, to detect the entire body, upper body, and lower body of a person in the image. It also determines the direction each of these body parts is facing. Finally, it combines these directional estimations to determine the overall direction the person is facing.
9. The object detector according to claim 8 , wherein, from a plurality of bins of each of the HOG features (B), the calculator selects the bin effective to obtain the feature pattern by a learning algorithm.
In the object detector where Histograms of Oriented Gradients (HOG) are used, and a classifier is used to identify an object in an image, a learning algorithm selects the most important gradient orientations (bins) from the pre-computed HOG features of the sample image. This optimizes the feature pattern used to train the classifier, so it's more effective in finding the target object. The calculator selects the bin effective to obtain the feature pattern by a learning algorithm.
10. The object detector according to claim 9 , wherein the learning algorithm is AdaBoost.
The object detector previously described, where Histograms of Oriented Gradients (HOG) and a classifier are used with a learning algorithm selecting bins, uses AdaBoost as the learning algorithm to select the most relevant gradient orientations (bins) from the pre-computed HOG features. This optimized bin selection results in a more accurate classifier for object detection.
11. The object detector according to claim 7 , wherein, from a plurality of bins of each of the HOG features (B), the calculator selects the bin effective to obtain the feature pattern by a learning algorithm.
In the object detector where Histograms of Oriented Gradients (HOG) are used to find objects, a learning algorithm selects the most important gradient orientations (bins) from the pre-computed HOG features of the sample image. This optimizes the feature pattern used to train the classifier, so it's more effective in finding the target object. The calculator selects the bin effective to obtain the feature pattern by a learning algorithm.
12. The object detector according to claim 11 , wherein the learning algorithm is AdaBoost.
This object detector identifies target objects within an image using Histograms of Oriented Gradients (HOG) features. A calculator first processes sample images to learn the object's characteristics. It computes multiple HOG features, each with varying numbers of bins, from different local areas of these sample images. To create a robust feature pattern for the object, the calculator selects the most effective bins from these HOG features using a **learning algorithm**. Specifically, this learning algorithm is **AdaBoost**. Once trained, the calculator then extracts HOG features (also with varying bin numbers) from local areas of a new 'target image'. A classifier, built from the learned feature pattern, uses these extracted features to judge the existence of the target object. ERROR (embedding): Error: Failed to save embedding: Could not find the 'embedding' column of 'patent_claims' in the schema cache
Unknown
December 9, 2014
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.